PyFFTW Playground

This repo contains recipes to install PyFFTW, and a python script to test the performnce of different ffts (numpy, scipy, FFTW) in pythob

Installation steps

Create environment:

make install_conda

Install pyfftw:

• option 1: local install, see LOCALINSTALL.md file
• option 2:

  pip install pyfftw
  

• option 3:

  conda install -c conda-forge pyfftw
  

Benchmarking

A python script called benchmark.py can be used to benchmark the speed of different ffts (numpy, scipy, FFTW) in your system. Run the script in the following way

python benchmark.py computer fft 10

where you can specify your computer name (eg mac_book_air_1_3_GHz_Intel_Core_i5), if you want to use numpy fft (fft) or fftpack (fftpack) - the first using MKL if installed with conda, the second being the slower if MKL is not available - and the number of times an fft is tested by timeit function (>10 to be somehow reliable).

This script will automatically create some summary plots, please commit them so with time we will get a library of performance tests and can see which fft is best in which environment.